A senior executive at Goldman Sachs has issued a stark warning regarding the rapid integration of artificial intelligence (AI) into the financial industry. Chris Churchman, a partner leading a flagship AI project for the firm, expressed concerns that Wall Street’s embrace of AI could potentially erode the critical thinking abilities of the next generation of finance professionals.
“There’s a huge risk here that we outsource our reasoning to these models, and we end up with cognitive atrophy, such that we can’t think from first principles ourselves anymore,” Churchman stated, as reported by CNBC. He is responsible for Goldman Sachs’ institutional client digital platform, Marquee.
Churchman drew a parallel between how modern technology has led to a decline in innate navigation and memory skills, suggesting that if algorithms handle all the heavy analytical lifting, bankers might gradually lose their own analytical prowess. “Reasoning still matters,” he emphasized. “You still need to think about the problem and organize it into a coherent argument, and now we’re handing off reasoning to AI.”
The push by Wall Street firms to deeply embed AI across trading and banking operations presents what could be a “devil’s bargain.” While AI promises short-term gains in profitability for the industry, it may, in the long run, undermine the very foundation of specialized talent it relies upon.
The Erosion of Foundational Training
As AI takes over routine tasks previously performed by junior bankers and traders, these early career professionals lose crucial opportunities to learn how to think and make decisions. This traditional learning-by-doing approach is being supplanted by AI, potentially sacrificing a vital career culture that cultivates junior staff into seasoned Wall Street talent.
This shift could even reduce the demand for junior bankers from the outset. Reports from last year indicated that Wall Street institutions were exploring ways to leverage AI to decrease the ratio of junior to senior staff. Churchman acknowledged the need for banks to strike a delicate balance between adopting AI and preserving the traditional “apprenticeship” style of training that has long been a hallmark of Wall Street.
Churchman, who previously headed UBS’s FX trading desk before joining Goldman Sachs in 2021, stated, “You need to learn by doing, and a lot of that knowledge is implicit, never written down.” He stressed the importance for Goldman Sachs to “ensure we don’t lose the implicit and intuitive knowledge that some of our best talent currently possesses, and at the same time ensure that the next generation can acquire those capabilities.”
Bridging the Gap Between Automation and Expertise
He offered an example: a junior trader handling client quote requests under the supervision of a senior trader. This process is instrumental in developing trading judgment. “We could absolutely automate that,” Churchman conceded, posing a critical question: “But would we then be able to produce seasoned traders who genuinely understand the business?”
When decisions involve high stakes and significant uncertainty, Churchman argued that systems must ensure employees retain ultimate decision-making authority, rather than becoming mere operators of AI systems. He also serves as co-chair of Goldman Sachs’ AI Working Group for Global Banking & Markets and admitted that even a premier investment bank like Goldman Sachs has not yet “fully figured out” how to manage this ongoing transformation.
AI in Practice: The Marquee Platform
During the podcast interview, Churchman also shared insights into Goldman Sachs’ experience integrating AI into its Marquee platform. Marquee provides institutional clients, such as hedge funds, with access to market data, research reports, risk analytics, and trade execution services.
Currently, the AI capabilities on Marquee are restricted to Goldman Sachs employees. From a technical standpoint, the primary challenge has been ensuring the AI’s outputs are 100% accurate and auditable. Churchman noted that while consumer-facing AI chatbots often include disclaimers about potential inaccuracies, the tolerance for error in high-stakes finance is exceptionally low.
He revealed a surprising self-assessment from the software during its development for a client-facing platform. “As we kept putting pressure on it and testing it rigorously, it was at least honest,” Churchman recounted. “It was like, ‘Look, at the end of the day, I’m better at sounding comprehensive than I am at being comprehensive.'”
